Accelerating method of Chinese handwriting recognition based on convolution neural network

The invention relates to an accelerating method of Chinese handwriting recognition based on a convolution neural network, which comprises the following steps: data preprocessing of handwritten character picture is convenient for classification and realization of convolution neural network; the structure of convolution neural network is constructed. The first classifier is added in the shallow layer, the second classifier is added in the middle layer, and the third classifier is added in the top layer. The method also comprises the steps of: initializing the network parameters, fixing the parameters of the shallow network and the first classifier, guiding the network training by using the loss function of the second classifier, and updating the parameters of the middle network and the second classifier; fixing the parameters of the shallow layer network, the middle layer network, the first classifier and the second classifier, guiding the network training by using the loss function of the third classifier, and updating the parameters of the top layer network and the third classifier; weighting the loss functions of the three classifiers, and fine-tuning the parameters of the whole network and the three classifiers by the weighted loss functions. Three classifiers of the network are tested on the test suite, the performance of the classifier is evaluated, and the computational complexity of the three classifiers is calculated.